►
Description
Chetan Surpur demonstrates a project he created that does anomaly detection on coordinates streaming out of the Unity Game Engine.
A
All
right,
hi,
everyone,
I'm,
Chayton,
I'm
gonna,
show
you
guys
a
platform
it's
so
unity
is
a
game
engine
where
you
can
create
physic
physical
simulations
and
it's
a
game,
physics
and
graphics
engine.
Do
you
create
a
physics
simulation
of
some
kind
of
some
kind
of
simulation
in
a
virtual
environment
and
what
I'll
show
you
today
is
a
an
API
into
unity
that
allows
you
to
get
your
data
coming
out
of
unity
and
into
new
pick.
So
you
can
do
anomaly:
detection
on
virtual
environments
that
you've
created
so
here's
a
this.
A
A
A
A
So,
as
you
drive
around
here,
so
I'm
controlling
it
right
now,
driving
the
car
and
you
see
that
the
tracer
that
leaves
behind
is
red.
So
it's
since
this
is
not
learned.
Yet
it's
a
new
pattern.
Everything
is
anomalous
and,
as
I
make
this
loop
a
couple
of
times,
you
should
start
sort
of
you'll
see
some
greens.
That's
new
pick
for
making
better
predictions
as
it
learns.
The
patterns
and
the
anomaly
scores
are
going
down.
A
A
That
was
soon
as
we
go
back
onto
the
path,
so
leaping
recognizes
it
and
it's
not
anomalous
anymore
and
if
I
can
even
drive
sort
of
the
opposite
direction
and
since
it's
a
temporal
sequence,
it's
it's.
You
know
it's
detecting
that
as
part
of
that
is
anomalous
so
anyway,
so
that's
the
anomaly
detection.
So
basically,
this
GPS
there's
like
a
sensor
on
this
car
GPS
sensor
that
it's
sending
its
data
to
this
Python
proxy
server
and
nupoc
is
basically
making
requests.
A
H-2B
request
to
that
processor
where
to
get
the
GPS
coordinates
visa
into
a
model,
returns
anomaly
score
to
the
proxy
server
which
represent
its
unity.
So
there's
this
there's
a
plug
in
here
at
this
github
repository
and
there's
instructions
on
how
sorry
you
can
get
you
just
import
unity
plug-in,
and
then
you
just
insert
lines
of
code
into
your
unity
application
to
send
a
those
new
pic
and
read
data
out
of
a
new
pic,
and
then
you
just
run
the
proxy
server
and
you
can
get
the
data.
A
So
here's
the
API
with
with
the
packet
beauty
package
you
can
import
and
then
this
all
this
code,
the
code
for
this
demo
I
just
showed
you
is
here
at
this
github
repo
and
you
can
use
that
as
a
starting.
You
know
a
starting
point
for
it.
If
you
want
to
do
a
hack
using
unity
because
it
has
a
unity,
new
pic
model
running
a
minute,
so
yeah
that's
available
there
on
get
home.
Yes,.
A
Yeah,
so
this
that's
the
question
about
the
geospatial
anomaly
demo,
and
so
the
question
is:
can
it
learn?
Multiple
patterns
are
multiple
like
routes,
and
and
yes
it
can
so
you
can,
you
can
I
could
have
drove
around
horizontally
instead
of
this
vertical
loop,
and
it
can
recognize
that
as
a
pattern
that
it
sees
anything
outside
of
those
two
would
be
anomalous,
so
it
can
definitely
learn
multiple
patterns
and
there's
more
I
mean
if
you
have
more
questions,
there's
a
whole.
We
did
a
geospatial
anomaly
detection
demo.
A
A
A
A
Multiple
cars
right
right,
so
you
could
I
did
well.
You
can
have
multiple
ones
and
the
interface
for
this
unity.
Api.
That
project
is
just
it
doesn't
really
care
what
the
inputs
and
outputs
are
you
just.
You
can
define
them
in
your
unity
project,
so
you
can
have
multiple
coordinates
coming
off
of
multiple
cars
going
into
a
new
pic
or
into
the
Python
side.
And
then
you
have
multiple
new
pic
models,
doing
anomaly:
detection
on
those
on
each
one
of
those
coordinates,
so
you
can
do
it
set
that
up.
There's
no
problem
with
that.